Unfolding the 3d swiss roll to a 2d manifold using VAE with perceptual loss.(sklearn.datasets.samples_generator.make_swiss_roll(n_points=5000, random_state=0, noise=0.5)